{"id":"W4413968810","doi":"10.2196/79010","title":"Automated Safety Plan Scoring in Outpatient Mental Health Settings Using Large Language Models: Exploratory Study","year":2025,"lang":"en","type":"article","venue":"JMIR Mental Health","topic":"Suicide and Self-Harm Studies","field":"Psychology","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Preprint; Mental health; Exploratory research; Plan (archaeology); Psychology; Medicine; Medical education; Computer science; Psychiatry; Geography; Sociology; World Wide Web","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01390357,0.000740228,0.0005128899,0.001017231,0.0005290366,0.001213836,0.001096698,0.0004545167,0.001254201],"category_scores_gemma":[0.07741071,0.0005703366,0.000784983,0.0009790428,0.0005388254,0.001013347,0.001548987,0.001074714,0.0003733661],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002115536,"about_ca_system_score_gemma":0.002767187,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02069112,"about_ca_topic_score_gemma":0.0277471,"domain_scores_codex":[0.9900548,0.007901075,0.000470245,0.0006153305,0.0007368932,0.0002215777],"domain_scores_gemma":[0.936289,0.05061875,0.005159757,0.003321679,0.003814594,0.0007962964],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001880915,0.008062383,0.8256586,0.0003806781,0.000349264,0.0003342594,0.01104399,0.01517225,0.001550154,0.0004229094,0.002784446,0.1323601],"study_design_scores_gemma":[0.0009002973,0.006409799,0.6320779,0.0002664773,0.0004159137,0.0004563199,0.01172337,0.3395439,0.003491198,0.001671912,0.002830595,0.0002123617],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9948311,0.00002688482,0.004026701,0.00007926614,0.000004151431,0.000399398,0.0003100177,0.00009200542,0.0002304321],"genre_scores_gemma":[0.9849454,0.00003842785,0.01356524,0.00004984785,0.000007335644,0.0005417939,0.000702313,0.00002189964,0.0001279047],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02069112,"threshold_uncertainty_score":0.07353002,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05361200590935531,"score_gpt":0.404078525044073,"score_spread":0.3504665191347177,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}